I was seeing a null pointer deref in fs/super.c:vfs_kern_mount().
Some file system get_sb() handler was returning NULL mnt_sb with
a non-negative return value. I also noticed a "hugetlbfs: Bad
mount option:" message in the log.
Turns out that hugetlbfs_parse_options() was not checking for an
empty option string after call to strsep(). On failure,
hugetlbfs_parse_options() returns 1. hugetlbfs_fill_super() just
passed this return code back up the call stack where
vfs_kern_mount() missed the error and proceeded with a NULL mnt_sb.
Apparently introduced by patch:
hugetlbfs-use-lib-parser-fix-docs.patch
The problem was exposed by this line in my fstab:
none /huge hugetlbfs defaults 0 0
It can also be demonstrated by invoking mount of hugetlbfs
directly with no options or a bogus option.
This patch:
1) adds the check for empty option to hugetlbfs_parse_options(),
2) enhances the error message to bracket any unrecognized
option with quotes ,
3) modifies hugetlbfs_parse_options() to return -EINVAL on any
unrecognized option,
4) adds a BUG_ON() to vfs_kern_mount() to catch any get_sb()
handler that returns a NULL mnt->mnt_sb with a return value
>= 0.
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
while ((p = strsep(&options, ",")) != NULL) {
int token;
+ if (!*p)
+ continue;
token = match_token(p, tokens, args);
switch (token) {
break;
default:
- printk(KERN_ERR "hugetlbfs: Bad mount option: %s\n", p);
- return 1;
+ printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
+ p);
+ return -EINVAL;
break;
}
}
config.gid = current->fsgid;
config.mode = 0755;
ret = hugetlbfs_parse_options(data, &config);
-
if (ret)
return ret;
error = type->get_sb(type, flags, name, data, mnt);
if (error < 0)
goto out_free_secdata;
+ BUG_ON(!mnt->mnt_sb);
error = security_sb_kern_mount(mnt->mnt_sb, secdata);
if (error)